ucb_agentic_ai

Lecture 09: Practical Lessons from Deploying Real-World AI Agents

Link to lecture recording on YouTube

Date: 2025-11-10

Speaker: Clay Bavor

Speaker’s Social Profile: Website / Google Scholar / GitHub / LinkedIn / X (Twitter)

Education:

Work:

Notes

Broadly categorize agents into 3 buckets:

Type Scope Examples
Personal agent trusted personal digital assistants ChatGPT, Gemini
Role-based / persona-based agent help get job done coding agents, legal agents
Customer-facing agent    

Analogy:

Speaker’s strong view: businesses transit from multiple channels (phone, chat, email etc.) to a single agent; AI architects inside of companies build, define and shape what the agent should look like

The conversation is the interface, no app to navigate, no hierarchical menu structure to get through

Sierra introduces new business model:

1997 era: the internet has been around for a couple of years, but no one had really figured out how to build and scale web services
Sierra: moving from agents as technology to agents as product, where agents can be configured, built, kept secure etc.

Today’s agents are transactional; the best agents do not resolve cases, they build relationships with a company’s customers across multiple interactions and transactions

Agents are non-deterministic for a given input, hence deserve an entirely new approach to software development

Build with vs. build or buy

Sierra provides platform-as-a-service, where lower level complexity is abstracted away; build once and deploy everywhere
every channel that a company interacts with its customers over is digital, and can be understood and broken down and attribute to what happened there

[Incomplete, work in progress]